Compare DER Across Clustering Definitions

Computes DER using different clustering (PSU) definitions and compares the results. This is useful for comparing, e.g., state-level vs PSU-level clustering to assess sensitivity of DER to the choice of primary sampling unit.

Usage

der_compare(x, clusters, ...)

Arguments

x

A draws matrix (S x d) or a svyder object. If a svyder object, the original draws and data are extracted automatically.

clusters

A named list of PSU vectors to compare. Each element should be an integer vector of length N.

...

Additional arguments passed to der_compute(). Required when x is a matrix: y, X, group, weights, family, sigma_theta, etc.

Value

A data.frame with columns: param, cluster_name, der.

See also

der_compute() for computing DER with a single clustering.

Other analysis: der_decompose(), der_sensitivity(), der_theorem_check()

Examples

data(nsece_demo)
# Compare DER using original PSU vs group-level clustering
comp <- der_compare(
  nsece_demo$draws,
  clusters = list(
    psu   = nsece_demo$psu,
    group = nsece_demo$group
  ),
  y = nsece_demo$y, X = nsece_demo$X,
  group = nsece_demo$group, weights = nsece_demo$weights,
  family = "binomial",
  sigma_theta = nsece_demo$sigma_theta,
  param_types = nsece_demo$param_types
)
head(comp)
#>      param cluster_name       der
#> 1  beta[1]          psu 0.2617696
#> 2  beta[2]          psu 2.6868825
#> 3  beta[3]          psu 0.3427294
#> 4 theta[1]          psu 3.3838218
#> 5 theta[2]          psu 0.6758803
#> 6 theta[3]          psu 1.1187639